Error-compensation network for ringing artifact reduction in holographic displays.

Area of Science:

• Optics and Photonics
• Computer Vision
• Machine Learning

Background:

• Learning-based computer-generated holography (CGH) has advanced phase-only hologram creation.
• Existing methods often overlook diffraction propagation models, leading to ringing artifacts and reduced image quality due to the Gibbs phenomenon.
• These artifacts stem from discrepancies between the learned hologram and the physical diffraction process.

Purpose of the Study:

• To develop a diffraction propagation error-compensation network for integration into existing CGH methods.
• To correct propagation errors by predicting residual values, thereby improving the accuracy of the diffraction process.
• To reduce the learning burden on CGH networks and mitigate ringing artifacts.

Main Methods:

• Proposed a diffraction propagation error-compensation network designed for seamless integration with current CGH algorithms.
• The network predicts residual errors to correct deviations in the diffraction propagation model.
• Evaluated the method through simulations and optical experiments on established CGH networks (HoloNet, CCNN).

Main Results:

• Achieved peak signal-to-noise ratios (PSNR) of up to 32.47 dB (HoloNet) and 29.53 dB (CCNN), outperforming baselines by 3.89 dB and 0.62 dB, respectively.
• Demonstrated a significant reduction in ringing artifacts in real-world holographic display experiments.
• The proposed compensation network effectively aligns the diffraction process with an ideal state.

Conclusions:

• The diffraction propagation error-compensation network successfully reduces artifacts and enhances image quality in CGH.
• This approach offers a flexible way to improve various CGH algorithms without requiring complete retraining.
• The method holds promise for advancing the quality and applicability of holographic display technologies.